# YOLOv8-M

> YOLOv8-M: one-stage object detector from the YOLOv8 family. Planned for LibreYOLO (not yet packaged).

- Family: YOLOv8 (External baseline)
- Parameters: 25.9M
- GFLOPs: 78.9
- Default input size: 640px
- Detection: anchor-free · nms
- Architecture: one-stage · CSPDarknet backbone
- Weight license: non-permissive
- Paper-reported mAP@50-95: 50.2%

## Benchmarks (COCO val2017)

| Hardware | Runtime | mAP@50-95 | FPS | Latency (ms) | VRAM (MB) |
|---|---|---|---|---|---|
| NVIDIA Jetson Orin Nano Super 8GB | ONNX Runtime FP32 | 49.4 | 1.4 | 740.5 | - |
| NVIDIA Jetson Orin Nano Super 8GB | PyTorch FP32 | 49.4 | 12.2 | 82.1 | 161 |
| NVIDIA Jetson Orin Nano Super 8GB | TensorRT FP16 | 49.4 | 24.6 | 40.6 | 14 |
| Raspberry Pi 5 | HailoRT INT8 | 42.4 | 26.7 | 37.4 | - |
| Raspberry Pi 5 | ncnn FP32 | 49.4 | 2.8 | 363.1 | - |
| Raspberry Pi 5 | ONNX Runtime FP32 | 49.4 | 1.0 | 972.3 | - |
| Raspberry Pi 5 | PyTorch FP32 | 49.4 | 1.1 | 906.5 | - |


Source: https://www.visionanalysis.org/model/yolov8m. Benchmarks produced with LibreYOLO (https://github.com/Libre-YOLO/libreyolo).